Research On Traffic Flow Forecasting Method Based On Ensemble Learning

In modern urban area,traffic conditions usually change rapidly.Road infrastructure and traffic resources in most cities are nearing saturation,and it is not possible to increase the infrastructure to alleviate traffic congestion.The increasing number of private cars is constantly challenging the transportation system at the same time.Using intelligent transportation system to discover the rules of traffic flow and predict the traffic flow to help the establishment of management and shunting measures has gradually become a new model of traffic management.The application of intelligent transportation systems can effectively alleviate traffic congestion and make full use of traffic resources.With the continuous development of technology,intelligent traffic systems collect traffic data through sensors and driving devices,making it possible to analyze traffic conditions and predict traffic flow dynamically.This paper uses ensemble learning to improve prediction accuracy.By constructing and combining multiple learning models to complete the learning task and integrating the prediction results,the prediction accuracy can be improved.Due to the influence of random factors in the data,different learning models will produce different prediction results.A stack integrated prediction model based on the extreme gradient boosting tree and the support vector regression is proposed.The extreme gradient boosting tree is used to transform the input of each record,and then use its prediction results to form new features.Finally,following optimization by particle swarm algorithm,support vector regression model is selected as the model for prediction.Related experiments on urban area data have proven that this integrated prediction method of regression and optimization is effective.Considering traffic flow will be affected by the road network structure and the spatial correlation,a traffic flow prediction method based on the integrated prediction of spatio-temporal data is proposed.The prediction result of this method is obtained by linear integration of the spatial data predicted by random forest and the time data predicted by Light GBM.The key parameters of the integrated model are adaptively adjusted by the genetic algorithm.Related experiments on traffic flow data on a certain road section show that the integrated model of spatio-temporal data has better prediction accuracy than using temporal data and spatial data for prediction alone.This article designs a prediction system based on traffic flow data.It can describe the traffic situation through a visual view of the traffic flow within a certain period of time,makes it easier to understand.Two ensembled models can be used to predict the traffic flow,and the prediction perfromance of the model can be observed through forecast errors and comparison charts.